Supply chain collaborative component replacement type selection method and system

Abstract

The invention relates to a supply chain collaborative component replacement type selection method and system. The method comprises the following steps: obtaining a candidate substitution set of the electronic component based on a dynamic knowledge graph; responding to a component state change event; according to the target application scene, calculating a matching degree between each component in the candidate replacement set and the technical requirement by using an adaptive weight model, and performing validity verification on a calculation result; integrating real-time supply chain data for collaborative optimization; and outputting an alternative type selection recommendation result after optimization sorting, and iteratively updating the dynamic knowledge graph and the self-adaptive weight model based on user feedback. According to the method, the candidate substitution set is updated in real time through the dynamic knowledge graph, scenarized parameter matching is realized in combination with the adaptive weight model, and multi-target optimization is performed by integrating the supply chain data, so that the problems of static graph updating delay, poor adaptability of the fixed weight model, supply chain data disjunction and the like are solved; the method has the advantages of improving the real-time performance and scene adaptability of the alternative scheme.

Description

Technical Field

[0001] This invention relates to the field of electronic component selection and substitution technology in supply chain management, specifically to a method and system for component substitution selection in supply chain collaboration. Background Technology

[0002] In the field of electronic component technology, the application of knowledge graph technology is becoming increasingly widespread, while the demand for supply chain collaborative optimization is also growing rapidly. Current electronic component replacement technologies mainly rely on structured data matching, and existing technical solutions have significant limitations.

[0003] In existing technologies, knowledge graph-based electronic component replacement methods typically employ static knowledge graphs. These graphs require manual maintenance and updates, and cannot respond in real time to dynamic changes in component status. For example, when a certain type of electronic component is discontinued, the system needs to manually mark and update replacement paths, leading to update delays. During this period, the generated replacement solutions may include discontinued components, severely impacting the feasibility of the replacement solutions.

[0004] Regarding parameter matching, existing technologies mostly use fixed-weight models to calculate parameter matching degree. This fixed-weight allocation strategy cannot adapt to the differentiated needs of different application scenarios. For example, in automotive electronics and consumer electronics scenarios, there are significant differences in the temperature resistance requirements of components, but existing systems use uniform weight calculation, resulting in a low compliance rate of recommended solutions in practical applications.

[0005] Furthermore, the supply chain data query function and parameter matching process in existing technologies are disconnected. Statistics show that a significant proportion of alternative solutions fail to be implemented because they do not adequately consider real-time supply chain conditions, such as sudden changes in delivery dates or insufficient inventory. This disconnect between technology and supply chain data severely impacts the practicality and feasibility of alternative solutions.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] In view of the above-mentioned deficiencies of the prior art, the first aspect of the present invention provides a component substitution selection method for supply chain collaboration, comprising the following steps: Step S1: Obtain a candidate replacement set for electronic components based on a dynamic knowledge graph; the dynamic knowledge graph is constructed and updated by parsing multi-source heterogeneous data, wherein the multi-source heterogeneous data includes at least: component parameter data, supply chain operation data, and design environment data; joint parsing of unstructured data is performed to extract key parameters; The unstructured data includes at least: PDF datasheets, circuit schematics, and BOM tables; the extracted key parameters include at least: structured parameters and their correlation with unstructured parameters. The structured parameters include at least electrical and physical characteristics; the unstructured associations are implicit relationships mined through graph attention networks. Step S2: In response to a component state change event; incrementally update the state of the corresponding node and / or the association between nodes in the dynamic knowledge graph, and resolve parameter conflicts based on confidence assessment; Step S3: Based on the target application scenario, use an adaptive weight model to calculate the matching degree between each component in the candidate alternative set and the technical requirements, and use a rule engine to verify the validity of the calculation results; Step S4: Integrate real-time supply chain data for collaborative optimization; use a multi-objective optimization model to jointly optimize and score the verified candidate solutions, wherein the score combines technical matching degree and supply chain stability under preset constraints; Step S5: Output the optimized and sorted alternative selection recommendation results, and iteratively update the dynamic knowledge graph and the adaptive weight model based on user feedback.

[0008] In the component replacement selection method for supply chain collaboration as described above, optionally, in step S2, the state change event includes active events and passive events; The active events are manufacturer EOL notifications and / or supply chain delivery date changes; the passive events are component call anomalies in EDA tool logs.

[0009] In the aforementioned component replacement selection method for supply chain collaboration, optionally, the sudden change in supply chain delivery time is defined as a fluctuation in supply chain delivery time exceeding 30%; and the abnormal component call in the EDA tool log is defined as three consecutive abnormal component call occurrences.

[0010] In the component substitution selection method for supply chain collaboration described above, optionally, incrementally updating the dynamic knowledge graph includes the following steps: The Temporal Graph Convolutional Network (TGCN) is used to dynamically predict the state of graph nodes. The calculation formula is as follows: ; in, This represents the embedding vector of node v at time t. Let v be the set of neighboring nodes. Let W and B be the real-time features of node v at time t, and let σ be the activation function. When parameter contradictions are detected, a confidence assessment based on Bayesian inference is used for conflict resolution, and its calculation formula is as follows: ; Among them, the posterior probability This represents the confidence level that the parameter takes the correct value under the observed data D; Likelihood This represents the conditional probability of observing data D when the parameter is the correct value; Prior probability This represents the initial probability that the value is correct based on historical experience before any observed data D is available; This represents a weighted summation of all candidate hypothesis values, used for normalization.

[0011] In the component substitution selection method for supply chain collaboration described above, optionally, in step S3, the adaptive weight model is configured to achieve dynamic weight allocation through a meta-learning framework and to be validated in conjunction with formal rules, including the following sub-steps: The input must include at least the scene feature vector. and environmental constraint vector The scene feature vector is an eight-dimensional vector obtained by mapping the application scene description through Word2Vec, and the environmental constraint vector is a six-dimensional vector obtained by quantifying data including at least temperature range, vibration conditions and electromagnetic compatibility level. The parameter weight matrix is ​​generated by the hierarchical gated recurrent unit (HGRU), and its calculation formula is as follows: ; The semicolon ";" indicates a vector concatenation operation, which concatenates scene feature vectors. and environmental constraint vector The concatenated input vectors form a higher-dimensional joint input; the hierarchical gated recurrent unit (HGRU) performs temporal modeling and feature extraction on the concatenated input vectors, capturing the coupling relationship between the application scenario and environmental conditions, and finally outputs a set of unnormalized weight representations; Softmax is used to convert an arbitrary real vector into a probability distribution, normalize the weights output by the HGRU, and make the sum of all parameter weight values ​​equal to 1; each dimension of the final parameter weight matrix W corresponds to the importance of a technical parameter, which includes at least temperature resistance, packaging compatibility, and electrical performance.

[0012] In the component replacement selection method for supply chain collaboration described above, optionally, the rule engine is a formal rule verification layer based on Datalog, used to define electronic constraints and industry standards. The rule execution logic is as follows: the candidate solution must pass all applicable rule verifications; otherwise, the adaptive weight model is triggered to redistribute weights or the candidate solution is eliminated.

[0013] In the component substitution selection method for supply chain collaboration as described above, optionally, in step S4, the supply chain collaboration optimization of the multi-objective optimization model includes the following steps: Step S4.1: Supply Chain Data Integration: Connect to the supplier database via RESTful API to obtain data including at least real-time inventory status, delivery date forecasts, and price volatility coefficients. Then, normalize and score this data using the following formula: ; ; Step S4.2: Construct a multi-objective optimization function to maximize the weighted sum of technology matching degree and supply chain stability. The objective function is: ; in, For technical compatibility, Rating the supply chain This is a scene adjustment coefficient, with a value of 0.7 in industrial scenarios and 0.5 in consumer electronics scenarios. Step S4.3: Apply preset constraints; the constraints are: technology matching degree. ≥0.8, delivery time score >0.6, inventory score >0.4; Step S4.4: Real-time filtering and sorting; the Pareto front screening algorithm is used to extract the non-dominated solution set, and the non-dominated solution with the highest technology matching degree is recommended first according to the output strategy, and the solution with the best supply chain stability is selected second.

[0014] To achieve the above objectives, a second aspect of the present invention provides a supply chain collaborative component substitution selection system, wherein the supply chain collaborative component substitution selection method as described in any of the embodiments of the first aspect includes: A dynamic knowledge graph management module is used to construct and dynamically update an electronic component knowledge graph based on multi-source heterogeneous data, and to obtain a candidate replacement set; the multi-source heterogeneous data includes at least component parameter data, supply chain operation data, and design environment data; this module includes: A multimodal data extraction unit is used to jointly parse unstructured data and extract key parameters; the unstructured data includes at least PDF datasheets, circuit schematics, and BOM tables; the extracted key parameters include at least structured parameters and unstructured relationships; the structured parameters include at least electrical characteristics and physical characteristics; the unstructured relationships are implicit relationships mined through graph attention networks; The incremental update and event response unit is used to respond to component state change events, incrementally update the state of the corresponding node and / or the association between nodes in the dynamic knowledge graph, and resolve parameter conflicts based on the confidence assessment of Bayesian inference. An adaptive similarity calculation engine is used to calculate the matching degree between each component and technical requirement in the candidate alternative set based on the target application scenario using a meta-learning-driven adaptive weight model, and to verify the effectiveness of the candidate solutions through an integrated rule engine. The supply chain collaborative optimization module is used to integrate real-time supply chain data and perform joint optimization scoring on verified candidate solutions through a multi-objective optimization model. The scoring combines technical matching degree and supply chain stability under preset constraints. The recommendation and feedback closed-loop module is used to output the optimized and ranked alternative selection recommendation results, and to iteratively update the dynamic knowledge graph and the adaptive weight model based on user feedback data.

[0015] To achieve the above objectives, a third aspect of the present invention provides a terminal device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when running the program, implements a component substitution selection method for supply chain collaboration as described in any of the preceding first aspects.

[0016] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions or a computer program, which, when processed and executed by a processor, implement the component substitution selection method for supply chain collaboration as described in any of the first aspects above.

[0017] The present invention provides a component substitution selection method and system for supply chain collaboration. It updates the candidate substitution set in real time through a dynamic knowledge graph, achieves scenario-based parameter matching by combining an adaptive weight model, and integrates supply chain data for multi-objective optimization. It solves the problems of static graph update delay, poor adaptability of fixed weight model and disconnection of supply chain data. It has the advantages of improving the real-time performance, scenario adaptability and implementation feasibility of substitution solutions.

[0018] Specifically, this invention achieves a significant leap forward in the accuracy, timeliness, and feasibility of electronic component replacement selection by constructing a dynamically updated knowledge graph, introducing a scenario-adaptive similarity calculation engine, and deeply integrating real-time supply chain data. The system can automatically capture and respond to market dynamics such as component production halts and parameter changes, ensuring that recommendations are always in sync with industry realities. Its innovative meta-learning weight model can automatically adjust parameter priorities according to different scenarios such as automotive electronics and industrial control, greatly improving the fit between solutions and scenario requirements. Simultaneously, by simultaneously verifying inventory, delivery time, and price fluctuations during the recommendation phase, it effectively avoids the problem of traditional solutions failing due to procurement infeasibility.

[0019] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an embodiment of a component substitution selection method for supply chain collaboration provided by the present invention. Figure 2 yes Figure 1 A flowchart illustrating the process when receiving an EOL notification from the manufacturer in step S2. Detailed Implementation

[0021] To make the technical means, inventive features, objectives, and effects of the invention readily understandable, the invention is further illustrated below with reference to specific figures. However, the invention is not limited to the embodiments described below.

[0022] It should be noted that the structures, proportions, sizes, etc., illustrated in the accompanying drawings of this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0023] Terms such as “comprising” and “including” indicate that, in addition to the components that are directly and explicitly stated in the specification and claims, the technical solution of the present invention does not exclude the presence of other components that are not directly or explicitly stated.

[0024] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0025] In traditional electronic component replacement selection systems, the static knowledge graph update mechanism relies on manual intervention, failing to capture component status change events in real time, resulting in replacement solutions including failed components. Fixed-weight models use uniform parameter evaluation standards, unable to dynamically adjust technical indicator priorities based on application scenarios, leading to insufficient matching of key performance parameters. Supply chain data is independent of the parameter matching process, lacking a real-time constraint mechanism for inventory and delivery date fluctuations on replacement solutions, thus reducing the feasibility of recommended solutions.

[0026] For example, in a component replacement scenario for automotive electronic control units, a certain model of IGBT module needs urgent replacement due to a manufacturer's discontinuation notice. The static knowledge graph, due to update delays, failed to promptly mark the device's status, and the system continued to recommend discontinued models. The fixed-weight model assigned fixed weights to temperature resistance indicators, failing to consider the impact of extreme temperature cycling on device reliability in this scenario, leading to performance degradation of the recommended devices during high-temperature testing. Furthermore, the supply chain module did not integrate real-time inventory data, resulting in actual inventory levels of replacement devices in the recommended solutions falling below safety thresholds, and procurement cycles exceeding the project's timeline.

[0027] If the above issues are not addressed, the alternative selection system will be unable to meet the technical requirements of high-reliability scenarios, leading to extended product development cycles and increased quality risks. The disconnect between supply chain data and parameter matching processes will increase the failure rate of recommended solutions, causing production line shutdowns due to material shortages. Delayed updates to the static knowledge graph will continuously generate invalid alternative paths, reducing system credibility and user reliance.

[0028] In response, this application proposes a component substitution selection method based on supply chain collaboration, which may include the following steps: Step S1: Obtain a candidate replacement set for electronic components based on a dynamic knowledge graph. The dynamic knowledge graph is constructed and updated by parsing multi-source heterogeneous data.

[0029] In an optional embodiment, the LayoutLMv3 multimodal model can be used to jointly parse unstructured data and extract key parameters.

[0030] The multi-source heterogeneous data can include component parameter data, supply chain operation data, and design environment data. Joint analysis of unstructured data is performed to extract key parameters. Unstructured data can include PDF datasheets, circuit schematics, and BOMs. Extracted key parameters can include structured parameters and unstructured correlations. Structured parameters can include electrical characteristics (e.g., withstand voltage, tolerance) and physical characteristics (e.g., package size, materials). Unstructured correlations are implicit relationships mined through graph attention networks (e.g., topology compatibility between discontinued and replacement MOSFET models).

[0031] The dynamic knowledge graph refers to a knowledge network built and updated in real time by continuously integrating multi-source heterogeneous data. In this application, a temporal graph convolutional network is used to dynamically predict node states and Bayesian inference is combined to resolve parameter conflicts. This enables timely reflection of component state changes and supply chain dynamics, solving the problem of alternative solutions failing due to the lag in updates of traditional static knowledge graphs. Multi-source heterogeneous data integration refers to simultaneously processing structured parameter data and unstructured design documents. In this application, a graph attention network can be used to mine implicit relationships in PDF manuals, breaking through the limitations of a single data source and improving the technical matching accuracy of alternative solutions.

[0032] In some of the solutions mentioned above in this application, the triggering mechanism for state change events lacks a clear classification standard, which makes it impossible for the system to accurately distinguish event types from different sources. This may cause delays or misjudgments in response to sudden changes in the supply chain or abnormalities in the design environment, thereby affecting the timeliness and accuracy of knowledge graph updates.

[0033] In this regard, this application further proposes that status change events include active events and passive events. Active events are vendor EOL notifications and / or supply chain delivery date changes, while passive events are component call anomalies in EDA tool logs.

[0034] In an optional embodiment, a sudden change in supply chain delivery time is defined as a fluctuation in supply chain delivery time exceeding 30%, and an abnormal component call in the EDA tool log is defined as three consecutive abnormal component call occurrences.

[0035] The threshold for supply chain delivery time fluctuations can be set based on industry benchmark test data or historical fluctuation distribution curves. For example, by statistically analyzing delivery time data from the past 12 months, the mean squared deviation can be calculated, and three standard deviations can be selected as the critical value. For determining three consecutive abnormal calls, a sliding time window mechanism can be used. For example, an event can be triggered if three consecutive call failures are detected within 30 seconds. This time window can be adjusted based on the system's response latency characteristics. The combined application of the delivery time fluctuation threshold and the number of abnormal calls forms a dual filtering mechanism. The former filters out regular fluctuations through amplitude screening, while the latter eliminates occasional interference through temporal continuity constraints.

[0036] Specifically, when supply chain delivery date data is detected, the system calculates the percentage deviation between the current delivery date and the historical benchmark in real time. If the deviation exceeds 30%, it is identified as a delivery date abrupt change event, triggering a supply chain status update for the corresponding component node in the knowledge graph. For monitoring EDA tool logs, the system uses a state machine model to track call records. When three consecutive call requests return error codes, the exception handling process is activated to update the component availability flag in the knowledge graph. This dual-judgment mechanism works in conjunction with the incremental update module of the knowledge graph. The delivery date fluctuation threshold ensures that only substantial supply chain risks are captured, and continuous anomaly detection effectively identifies persistent design problems. By embedding quantitative standards into the event detection algorithm, the system maintains a high response speed while keeping the false alarm rate below 5%, significantly improving the reliability of alternative selection decisions.

[0037] Active events are configured to receive EOL notification messages directly from manufacturers via API interfaces or capture sudden changes in delivery time data through the supply chain management system. Delivery time fluctuations can be quantified as triggering an event when the delivery time fluctuates by more than 30%. For example, if the delivery time of a certain type of resistor is extended from 7 days to 10 days, the system determines that the threshold has not been reached; however, if it is extended to 30 days, an event is triggered. Passive events are implemented by real-time parsing of log files generated by EDA tools. An event is triggered when the same component experiences three consecutive call anomalies within 24 hours. Anomaly types include schematic symbol loading failure, SPICE model simulation errors, or design rule check mismatches.

[0038] Specifically, when an active event is detected, the manufacturer's EOL notification extracts component model, discontinuation date, and alternative recommendations by parsing the XML-formatted production stoppage announcement file, triggering an update of the status marker of the corresponding node in the knowledge graph. Supply chain delivery time anomalies are detected by real-time monitoring of the delivery time field in the supplier database; when the increase exceeds a set threshold, the relationship weights between inventory nodes and supplier nodes in the knowledge graph are automatically adjusted. Passive events are detected by listening to JSON-formatted log streams generated by EDA tools, using regular expressions to match preset error code patterns, such as "ERR_COMPONENT_LOAD_FAILURE" or "WARN_DRC_VIOLATION," identifying abnormal components, and triggering a reassessment of the confidence level of the technical parameter nodes in the knowledge graph.

[0039] In this embodiment, the two event types are transmitted through independent message queue channels. Active events use a high-priority queue to ensure millisecond-level response, while passive events use a batch processing queue for minute-level aggregation analysis. When both active and passive events point to the same component, the incremental update operation is configured to prioritize the state change triggered by the active event, followed by the superimposed correction of technical parameters detected by the passive event. For example, after a capacitor is marked as discontinued, its failure records in different circuit designs will further verify the accuracy of the state change.

[0040] Through the above technical solution, this application achieves accurate classification and rapid response to component status change events. As a result, the system can promptly capture active signals in the supply chain and passive anomalies in the design environment, thereby improving the accuracy and real-time performance of knowledge graph updates. Specifically, for active events, the system can immediately detect changes in the component lifecycle or supply chain fluctuations, and then quickly adjust alternative strategies. For passive events, the system can identify potential problems from actual design applications, avoiding the recommendation of inapplicable alternatives. This dual event monitoring mechanism significantly reduces the delay in knowledge graph updates while improving the comprehensiveness of the updated content. Furthermore, by distinguishing between active and passive events, the system can adopt differentiated response strategies for different types of changes, such as prioritizing active events and conducting in-depth analysis before updating passive events. This approach not only improves the system's ability to respond to sudden changes in the supply chain but also enhances the accuracy of identifying anomalies in the design environment, thereby ensuring the reliability and applicability of component replacement selection.

[0041] Step S2: In response to component state change events, incrementally update the state of the corresponding node and / or the relationships between nodes in the dynamic knowledge graph, and resolve parameter conflicts based on confidence assessment.

[0042] like Figure 2 As shown, when the system receives an EOL (End of Life) notification from the manufacturer, this is the starting point of the process, indicating that the manufacturer has issued an EOL notification for the product. Next, the system parses key fields from the notification. This may include information such as the product model and the date support ends. Then, the system updates the status of relevant graph nodes. This step ensures that the information in the system is up-to-date and reflects the product's EOL status. Finally, the system triggers an alternative path search based on the updated status. This may involve continuing to search for alternative products or solutions so that users can still receive service after the original product ceases support.

[0043] In this step, incrementally updating the dynamic knowledge graph can further include the following steps: The Temporal Graph Convolutional Network (TGCN) is used to dynamically predict the state of graph nodes. The calculation formula is as follows: ; in, This represents the embedding vector of node v at time t. Let v be the set of neighboring nodes. Let W represent the real-time features of node v at time t, and let W and B be the weight parameters to be trained, and σ be the activation function.

[0044] When parameter contradictions are detected, a confidence assessment based on Bayesian inference is used for conflict resolution, and its calculation formula is as follows: ; Among them, the posterior probability This represents the confidence level that the parameter takes the correct value under the observed data D; Likelihood This represents the conditional probability of observing data D when the parameter is the correct value; Prior probability This represents the initial probability that the value is correct based on historical experience before any observed data D is available; This represents a weighted summation of all candidate hypothesis values, used for normalization.

[0045] Optionally, in the dynamic prediction step of the temporal graph convolutional network, the node embedding vector is generated by fusing historical states, neighbor node states, and real-time features. The weight parameters are optimized during training using the backpropagation algorithm, and the LeakyReLU activation function can be used to alleviate the gradient vanishing problem. In the confidence evaluation step of Bayesian inference, the prior probability is obtained based on historical data statistics. For example, if the withstand voltage of a certain type of capacitor matches the manufacturer's nominal value in 93 out of the past 100 updates, then its prior probability is set to 0.93. The normalization term can be approximated using the Monte Carlo sampling method, and the candidate hypothesis value set contains parameter values ​​provided by at least three data sources.

[0046] Specifically, when a sudden change in supply chain delivery dates triggers a knowledge graph update, the temporal graph convolutional network first generates a new predicted node state value based on the node's embedding vectors from the past five time steps, the inventory status of adjacent supplier nodes, and the real-time updated delivery date data. If a deviation of more than 15% is detected between the temperature range parameter in the manufacturer's EOL notification and the measured value recorded in the EDA tool logs, a Bayesian conflict resolution mechanism is activated: the joint probability of each candidate parameter value is calculated, with the prior probability of the manufacturer's data set at 0.85 and the prior probability of the EDA log data set at 0.65. Finally, the parameter value with a posterior probability greater than 0.8 is selected as the basis for the update. This synergistic effect of temporal modeling and probabilistic inference reduces the node state update delay from 6-8 hours in traditional methods to within 30 minutes, and improves the parameter conflict resolution accuracy to over 92%.

[0047] For example, when conflicting values ​​are obtained from different data sources for the rated power parameter of a certain type of resistor (e.g., the datasheet shows 1W, while the measured data is 0.9W), the reliability of each data source can be evaluated by calculating the posterior probability. Assuming the prior probability of the datasheet is 0.8 and the prior probability of the measured data is 0.6, the updated confidence level can be obtained through Bayesian inference, thus selecting the most reliable parameter value for knowledge graph updates.

[0048] Through the above technical solutions, this application can effectively capture the dynamic characteristics of component parameters evolving over time, improving the real-time performance and accuracy of knowledge graph updates. The dynamic prediction method based on temporal graph convolutional networks can integrate historical node states, neighbor information, and real-time features, thereby more accurately predicting future state changes of components. Simultaneously, the confidence assessment mechanism using Bayesian inference can effectively resolve parameter conflicts when parameter contradictions arise from multi-source heterogeneous data by quantitatively evaluating the reliability of different data sources, improving the efficiency and accuracy of node state updates. This method is not only applicable to conflict resolution for single parameters but can also be extended to complex scenarios with multiple parameters and multiple data sources, providing reliable technical support for the maintenance and updating of dynamic knowledge graphs.

[0049] Step S3: Based on the target application scenario, use an adaptive weight model to calculate the matching degree between each component in the candidate alternative set and the technical requirements, and use a rule engine to verify the validity of the calculation results.

[0050] In step S3, the adaptive weight model is configured to achieve dynamic weight allocation through a meta-learning framework and is validated using formal rules. This may further include the following sub-steps: The input must include at least the scene feature vector. and environmental constraint vector The scene feature vector is an eight-dimensional vector obtained by mapping the application scene description through Word2Vec, and the environmental constraint vector is a six-dimensional vector obtained by quantifying data including at least temperature range, vibration conditions and electromagnetic compatibility level.

[0051] The parameter weight matrix is ​​generated by the hierarchical gated recurrent unit (HGRU), and its calculation formula is as follows: ; The semicolon ";" indicates a vector concatenation operation, which concatenates scene feature vectors. and environmental constraint vector The concatenated input vectors form a higher-dimensional joint input. The Hierarchical Gated Recurrent Unit (HGRU) performs temporal modeling and feature extraction on the concatenated input vector, capturing the coupling relationship between the application scenario and environmental conditions. The final output is a set of unnormalized weights. Softmax is used to transform an arbitrary real-valued vector into a probability distribution, normalizing the weights output by the HGRU so that the sum of all parameter weights is 1. Each dimension of the final parameter weight matrix W corresponds to the importance of a technical parameter, which may include temperature resistance, packaging compatibility, electrical performance, etc.

[0052] Therefore, each dimension of the resulting parameter weight matrix W corresponds to the importance of a certain technical parameter, including temperature resistance, package compatibility, and electrical performance. For example, in automotive electronics applications, the model might output a weight of 0.4 for temperature resistance, 0.3 for package compatibility, and 0.3 for electrical performance. In consumer electronics, the weight allocation might be adjusted to 0.2 for temperature resistance, 0.4 for package compatibility, and 0.4 for electrical performance. In industrial control applications, the weight for temperature resistance increases from the baseline value of 0.3 to 0.6, and the weight for package compatibility increases from 0.2 to 0.4.

[0053] The scene feature vector maps the application scene described in natural language into an eight-dimensional numerical vector using the Word2Vec model. For example, "automotive engine control unit" is mapped as [0.32, -0.15, 0.78, 0.04, -0.23, 0.56, 0.11, -0.09]. The environmental constraint vector generates a six-dimensional vector by quantizing the temperature range, vibration conditions, and electromagnetic compatibility level. For example, the temperature range of -40℃ to 150℃ is encoded as [0.8, 0.6, 0.7, 0.0, 0.0, 0.0]. The Hierarchical Gated Recurrent Unit (HGRU) adopts a two-layer gating structure. The first layer processes the temporal correlation of the scene feature vector, and the second layer integrates the dynamic changes of the environmental constraint vector, achieving cross-layer feature interaction through hidden state propagation. In the weight normalization process, the Softmax function converts the real-valued vector output by the HGRU into a probability distribution. For example, it transforms the original output [2.3, -1.5, 0.8] into [0.85, 0.05, 0.10], ensuring that the temperature resistance weight reaches 0.85 in the automotive electronics scenario. The parameter weight matrix is ​​multiplied with the component parameters in the dynamic knowledge graph, automatically adapting to the current scenario requirements when generating the matching score.

[0054] Specifically, the application scenario description text is first converted into an eight-dimensional feature vector, and the environmental constraint data is quantized into a six-dimensional vector. The two are then concatenated to form a fourteen-dimensional joint input vector. The Hierarchical Gated Recurrent Unit (HGRU) analyzes the dynamic correlation between scenario features and environmental conditions through time-series modeling, such as automatically enhancing the weight of temperature resistance parameters in scenarios with sudden temperature changes. The generated unnormalized weights are processed by Softmax to form a standardized weight matrix, ensuring that the weight allocation of different technical parameters conforms to probability distribution constraints. When the rule engine detects that the electrical performance weight is below 0.2, it triggers a weight reassignment mechanism to readjust the parameter importance. The final output weight matrix is ​​weighted and calculated with the technical parameters of candidate components. Combined with real-time data from the supply chain collaborative optimization module, this achieves joint optimization of technology matching and supply chain stability. This solution continuously optimizes the HGRU's weight generation strategy through a meta-learning framework, enabling parameter weight allocation to adapt to different combinations of application scenarios and environmental conditions, effectively improving the accuracy of alternative selection matching.

[0055] Through the above technical solution, this application achieves accurate matching between the importance assessment of technical parameters and specific application scenarios and environmental conditions. The adaptive weight model can dynamically adjust parameter weights according to different scenarios, overcoming the limitations of traditional fixed-weight models in differentiated application environments. This improves the calculation accuracy of the alternative selection matching degree, making the recommended alternatives more in line with actual application needs. Simultaneously, by capturing the coupling relationship between scenario features and environmental constraints, the model's adaptability to complex application environments is enhanced. Furthermore, formal rule verification further ensures the rationality and effectiveness of weight allocation.

[0056] In this embodiment, optionally, the rule engine can be a formal rule validation layer based on Datalog, used to define electronic constraints and industry standards. The rule execution logic is as follows: a candidate solution must pass all applicable rule validations; otherwise, the adaptive weight model is triggered to redistribute weights or the candidate solution is eliminated.

[0057] The formal rule verification layer defines inviolable rigid constraints using the Datalog language. These constraints can include electrical parameter thresholds and physical compatibility standards. For example, the withstand voltage must be greater than or equal to 1.2 times the circuit design value, and the pad size tolerance must meet the ±0.15mm range specified in the IPC-7351 standard. The rule execution logic employs a short-circuit verification mechanism, immediately terminating the subsequent verification process when any rule verification fails. Weight redistribution adjusts the weight matrix output by the hierarchical gated loop unit, reducing the weight of parameters that failed verification to 30%-50% of their original value while increasing the weight proportion of related parameters. The candidate solution elimination criterion is set as the presence of rule verification failures after three consecutive weight adjustments.

[0058] Specifically, when a candidate solution enters the rule verification layer, it first loads the set of constraint rules associated with the current application scenario. For example, in the automotive electronics scenario, the 23 verification rules corresponding to the AEC-Q100 reliability standard are automatically loaded. The rule engine executes the logical judgment of all applicable rules through parallel computation. When a withstand voltage value lower than 25V or a package size exceeding the allowable range of PCB layout is detected, a verification failure flag is immediately generated. For candidate solutions that fail verification, a hierarchical gated loop unit is triggered to recalculate the parameter weight matrix, compressing the weight values ​​of the failed parameters to 40% of their original values, while increasing the weights of associated parameters by 10%-15%. If all constraint rules still cannot be met after three iterations, the solution is removed from the candidate set. This processing mechanism ensures that the final output recommended solution simultaneously meets the requirements of dynamic matching optimization and engineering feasibility. For example, in the operational amplifier selection scenario, it effectively excludes candidate devices that meet the gain-bandwidth product but whose supply voltage range does not meet the design specifications.

[0059] Through the above technical solution, this application achieves rigid constraints on the output results of the adaptive weight model. The rule engine transforms abstract engineering specifications into quantifiable verification rules, ensuring that the recommendation results meet both the matching degree requirements calculated by the dynamic weight model and the uncompromising hard technical specifications. When a rule verification fails, the weight reassignment mechanism of the adaptive weight model is triggered, and the matching degree is recalculated by adjusting the parameter weights, attempting to find a suboptimal solution within the rule framework; if it still fails, the candidate solution is directly eliminated to avoid outputting recommendation results with fundamental flaws. This dual processing mechanism maintains the flexibility of the weight allocation model while ensuring the feasibility of the technical solution through formal rule verification, effectively solving the problem that the technical matching degree calculation conforms to the weight allocation logic but violates the fundamental principles of electronics or mandatory industry standards.

[0060] Step S4: Integrate real-time supply chain data for collaborative optimization. A multi-objective optimization model is used to jointly optimize and score the validated candidate solutions. The score considers both technical matching and supply chain stability while meeting preset constraints.

[0061] Optionally, in step S4, the supply chain collaborative optimization of the multi-objective optimization model may specifically include the following steps: Step S4.1: Supply Chain Data Integration: Connect to the supplier database via RESTful API to obtain data including at least real-time inventory status, delivery date forecasts, and price volatility coefficients. Then, normalize and score this data using the following formula: ; .

[0062] Step S4.2: Construct a multi-objective optimization function to maximize the weighted sum of technology matching degree and supply chain stability. The objective function is: ; in, For technical compatibility, Rating the supply chain This is a scene adjustment coefficient, with a value of 0.7 in industrial scenarios and 0.5 in consumer electronics scenarios.

[0063] Step S4.3: Apply preset constraints. The constraints are designed as: technical matching degree. ≥0.8, delivery time score >0.6, inventory score >0.4.

[0064] Step S4.4: Real-time filtering and sorting. The Pareto front screening algorithm is used to extract the non-dominated solution set, and the non-dominated solution with the highest technology matching degree is recommended first according to the output strategy, and the solution with the best supply chain stability is selected second.

[0065] Supply chain data integration can be achieved by connecting to supplier databases via RESTful APIs to obtain real-time inventory status, delivery date forecasts, and price fluctuation coefficients. Data normalization can be performed using linear transformations or probability distribution mappings. For example, delivery date scoring is achieved through exponential smoothing of the ratio of the delivery date forecast to the benchmark delivery date, while inventory scoring is achieved through piecewise linear transformation of the difference between the current inventory level and the safety stock threshold. When constructing the multi-objective optimization function, the weighting of technology matching degree and supply chain score in the weighted sum calculation can be dynamically adjusted based on the scenario type. In industrial scenarios, the weight of technology matching degree is set to 0.7, while in consumer electronics scenarios it is set to 0.5. This weight difference is automatically triggered by scenario feature vector identification. Preset constraints are set as a lower limit of 0.8 for technology matching degree, 0.6 for delivery date score, and 0.4 for inventory score, filtering candidate solutions that do not meet basic business requirements through hard thresholds. The Pareto front screening algorithm can be implemented using fast non-dominated sorting. After the non-dominated solution set is generated, a two-level recommendation strategy is set: the first level is sorted in descending order by technology matching degree, and the second level is sorted in descending order by supply chain score.

[0066] In other alternative embodiments, the supplier database can be Digi-Key, Mouser, Avnet, etc., the inventory status can be real-time inventory, in-transit quantity, etc., the delivery date forecast can be a dynamically adjusted value based on the LSTM model, and the price volatility coefficient can be obtained by calculating the ratio of the 30-day price standard deviation to the mean.

[0067] Specifically, the supply chain data integration step obtains dynamic indicators from the supplier database in real time through standardized interfaces. Delivery date forecasts are converted into delivery date scores within a 0-1 range using an exponential smoothing formula, and inventory status is mapped to inventory scores using a piecewise function, ensuring comparability of data across different dimensions. When constructing the multi-objective optimization function, the technology matching degree is calculated in the preceding steps. The supply chain score integrates delivery date scores, inventory scores, price fluctuation coefficients, and scenario adjustment coefficients through a weighted average. Preset values ​​are automatically matched based on the application scenario type. Industrial scenarios prioritize technical performance, while consumer electronics scenarios balance supply chain flexibility. When applying constraints, candidate solutions must simultaneously meet both the lower bound of technical matching and the lower bound of supply chain indicators to avoid achieving the standard in one dimension while rendering the whole solution infeasible. In the real-time filtering stage, the Pareto front algorithm is used to select non-dominated solutions from the candidate set, prioritizing the solution with the highest technical matching as the primary recommendation. When the primary recommendation fails due to supply chain fluctuations, it automatically switches to the secondary solution with the best supply chain stability. This dual-guarantee mechanism improves the success rate of the recommendation system.

[0068] Through the above technical solution, this application achieves dynamic integration of supply chain data and parameter matching. Real-time acquisition of key supply chain indicators via a RESTful API ensures data timeliness. A multi-objective optimization model incorporates technology matching degree and supply chain stability into a unified scoring system, achieving comprehensive evaluation. Preset constraints and Pareto front screening further guarantee the feasibility of the recommended solution. Therefore, this solution effectively solves the problem of disconnect between supply chain data and parameter matching, improving the accuracy and feasibility of component substitution selection.

[0069] Step S5: Output the optimized and ranked alternative selection recommendations, and iteratively update the dynamic knowledge graph and adaptive weight model based on user feedback.

[0070] In practice, the system outputs the optimized and ranked alternative selection recommendations to the user and records the user's acceptance or rejection of the recommendations through the interface. Subsequently, the system initiates an iterative update process based on this feedback data: on the one hand, it uses the feedback data to strengthen or weaken the substitution relationships between component nodes in the dynamic knowledge graph; on the other hand, it uses the user's actual choices as a supervision signal to fine-tune the adaptive weight model, making it more in line with the real selection preferences in different application scenarios, thereby achieving self-optimization and continuous learning of the recommendation system.

[0071] Through the above-described scheme, this application achieves a significant improvement in the real-time performance, accuracy, and reliability of electronic component replacement selection. The incremental update mechanism of the dynamic knowledge graph enables the system to respond promptly to changes in component status, effectively reducing the risk of replacement failures. The adaptive weight model combined with the rule engine's dual verification mechanism overcomes the limitations of the fixed weight model in different application scenarios, improving the technical matching degree of replacement solutions. The supply chain collaborative optimization module integrates real-time supply chain data into the decision-making process, significantly improving the feasibility of recommended solutions. The closed-loop feedback mechanism ensures that the system can continuously learn and optimize, adapting to the dynamically changing market environment. This comprehensive optimization scheme effectively solves the problems of lagging knowledge graph updates, fixed parameter weights, and disconnect between supply chain data and parameter matching in existing technologies, providing more efficient and reliable component replacement selection support for electronic product design and production.

[0072] Practical verification shows that this solution achieves precision, adaptability, and feasibility in electronic component substitution decisions through the triple synergy of the dynamic evolution mechanism of knowledge graphs, the weight allocation strategy driven by meta-learning, and the multi-objective supply chain optimization model. Compared with traditional methods, it achieves significant improvements in core indicators such as the first recommendation pass rate (+31%) and the solution implementation cycle (-65%).

[0073] To achieve the above objectives, the present invention also provides a component replacement selection system for supply chain collaboration, including a dynamic knowledge graph management module, an adaptive similarity calculation engine, a supply chain collaboration optimization module, and a recommendation and feedback closed-loop module.

[0074] The dynamic knowledge graph management module is configured to jointly parse unstructured data using a multimodal data extraction unit, extracting structured parameters and their associations with unstructured data. For example, it uses a graph attention network to mine implicit relationships in the functional coupling of circuit schematics, mapping electrical characteristic parameters to application scenario data in the BOM. The incremental update and event response unit is configured to handle parameter conflicts using a Bayesian inference-based confidence assessment mechanism. When a supply chain delivery time fluctuation exceeds 30% or three consecutive abnormal component calls are detected in the EDA tool logs, the state correction of knowledge graph nodes is triggered.

[0075] The adaptive similarity calculation engine is configured to generate a parameter weight matrix using a meta-learning-driven weight model. The hierarchical gated recurrent unit concatenates the scene feature vector and environmental constraint vector, outputting normalized weights through a Softmax function. The integrated rule validation layer is configured to forcibly exclude candidate solutions that violate electronic constraints based on Datalog formal rules. When encapsulation incompatibility is detected, weight reallocation or solution elimination is triggered. The adaptive similarity calculation engine inputs the scene feature vector and environmental constraint vector into the hierarchical gated recurrent unit to generate a parameter weight matrix adapted to different application scenarios.

[0076] The supply chain collaboration optimization module is configured to obtain delivery forecasts and inventory status data from the supplier database in real time via a RESTful API, and uses normalized scoring and Pareto front screening algorithms in the multi-objective optimization model.

[0077] Specifically, the multimodal data extraction unit in the dynamic knowledge graph management module extracts electrical characteristic parameters and functional coupling relationships by jointly parsing PDF datasheets and circuit schematics, forming a multi-source heterogeneous knowledge graph containing component parameter data, supply chain operation data, and design environment data. When the incremental update and event response unit receives an EOL notification from the manufacturer, it triggers a time-series graph convolutional network to dynamically predict the state of the graph nodes and calculates the confidence of the updated nodes based on the posterior probability of Bayesian inference.

[0078] The recommendation and feedback closed-loop module uses data of alternative solutions actually adopted by users as training samples and iteratively updates the parameter configuration of the graph attention network through the backpropagation algorithm. For example, when a recommendation result is not adopted, the system automatically adjusts the confidence threshold of unstructured associations and recalculates the matching degree ranking of candidate solutions. This achieves closed-loop linkage between knowledge graph updates, parameter matching calculations, and supply chain optimization, effectively solving the problem of imbalance in technical indicators and supply chain stability caused by the fragmentation of collaboration between modules.

[0079] The specific implementation methods have been described in detail above and will not be repeated here. Through the above technical solutions, this application achieves closed-loop linkage between knowledge graphs and supply chain data, solving the problem of fragmented collaboration between modules leading to the inability to optimize technical indicators and supply chain stability in real time. The incremental update mechanism of the dynamic knowledge graph ensures timely response to component status change events, avoiding the failure of alternative solutions due to data lag. The adaptive weight model improves parameter matching accuracy through scenario-driven dynamic allocation, and combines the mandatory verification of the rule engine to eliminate candidate solutions that violate electronic constraints. The multi-objective optimization model, by integrating real-time supply chain data and a normalized scoring mechanism, prioritizes the solution with the lowest supply chain risk based on technical feasibility. The feedback closed-loop mechanism continuously optimizes the model through user behavior data, forming an end-to-end self-learning system, significantly improving the reliability and implementation efficiency of alternative selection decisions.

[0080] In summary, this invention, through a real-time incremental update mechanism of a dynamic knowledge graph, can automatically capture key information such as component discontinuation and parameter changes, ensuring that recommendations are always synchronized with market dynamics. It innovatively introduces a meta-learning-driven adaptive weight allocation model, which can automatically adjust parameter priorities according to different application scenarios, significantly improving the fit between alternative solutions and scenario requirements. By deeply integrating real-time supply chain data, the system simultaneously verifies inventory levels, delivery date fluctuations, and price trends during the parameter matching stage, effectively avoiding solution failures due to procurement infeasibility. The system's built-in rule engine can automatically verify electronic constraints, avoiding technical risks caused by human error; through the linkage of a historical alternative case library and real-time data, the solution reliability verification time is shortened from the traditional 3 days to 15 minutes; combined with a supply chain scoring model, it quantitatively assesses inventory stability and cost fluctuations, providing an intuitive basis for procurement decisions. Practical applications show that after adopting this system, the maintenance cycle of industrial equipment is shortened by 42%, component procurement costs are reduced by 18%, and the first-time implementation success rate of alternative solutions increases to 89%, significantly reducing the risks of rework and inventory backlog.

[0081] To achieve the above objectives, the present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the program, it can implement the steps of the component substitution selection method for supply chain collaboration as described in any of the foregoing embodiments.

[0082] Processors and memory can be configured separately or integrated together, for example, integrated on a system-on-chip (SOC) in a terminal device.

[0083] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing computer-executable instructions or computer programs, which, when processed and executed, implement the component substitution selection method for supply chain collaboration as described above.

[0084] The computer-readable storage medium is, for example, memory. Memory can be volatile or non-volatile, or it can include both volatile and non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM).

[0085] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0086] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A component substitution selection method based on supply chain collaboration, characterized in that, Includes the following steps: Step S1: Obtain a candidate replacement set for electronic components based on a dynamic knowledge graph; The dynamic knowledge graph is constructed and updated by parsing multi-source heterogeneous data, which includes at least: component parameter data, supply chain operation data, and design environment data; unstructured data is jointly parsed to extract key parameters; The unstructured data includes at least: PDF datasheets, circuit schematics, and BOM tables; the extracted key parameters include at least: structured parameters and their correlation with unstructured parameters. The structured parameters include at least electrical and physical characteristics; the unstructured associations are implicit relationships mined through graph attention networks. Step S2: In response to a component state change event; incrementally update the state of the corresponding node and / or the association between nodes in the dynamic knowledge graph, and resolve parameter conflicts based on confidence assessment; Step S3: Based on the target application scenario, use an adaptive weight model to calculate the matching degree between each component in the candidate alternative set and the technical requirements, and use a rule engine to verify the validity of the calculation results; Step S4: Integrate real-time supply chain data for collaborative optimization; use a multi-objective optimization model to jointly optimize and score the verified candidate solutions, wherein the score combines technical matching degree and supply chain stability under preset constraints; Step S5: Output the optimized and sorted alternative selection recommendation results, and iteratively update the dynamic knowledge graph and the adaptive weight model based on user feedback.

2. The component replacement selection method according to claim 1, characterized in that, In step S2, the state change event includes active events and passive events; The active events are manufacturer EOL notifications and / or supply chain delivery date changes; the passive events are component call anomalies in EDA tool logs.

3. The component replacement selection method according to claim 2, characterized in that, The supply chain delivery time abrupt change is defined as a fluctuation in supply chain delivery time exceeding 30%; the component call anomaly in the EDA tool log is defined as three consecutive component call anomalies.

4. The component replacement selection method according to claim 2, characterized in that, Incremental updating of the dynamic knowledge graph includes the following steps: The Temporal Graph Convolutional Network (TGCN) is used to dynamically predict the state of graph nodes. The calculation formula is as follows: ； in, This represents the embedding vector of node v at time t. Let v be the set of neighboring nodes. Let W and B be the real-time features of node v at time t, and let σ be the activation function. When parameter contradictions are detected, a confidence assessment based on Bayesian inference is used for conflict resolution, and its calculation formula is as follows: ； Among them, the posterior probability This represents the confidence level that the parameter takes the correct value under the observed data D; Likelihood This represents the conditional probability of observing data D when the parameter is the correct value; Prior probability This represents the initial probability that the value is correct based on historical experience before any observed data D is available; This represents a weighted summation of all candidate hypothesis values, used for normalization.

5. The component replacement selection method according to claim 1, characterized in that, In step S3, the adaptive weight model is configured to achieve dynamic weight allocation through a meta-learning framework and to be validated using formal rules, including the following sub-steps: The input must include at least the scene feature vector. and environmental constraint vector The scene feature vector is an eight-dimensional vector obtained by mapping the application scene description through Word2Vec, and the environmental constraint vector is a six-dimensional vector obtained by quantifying data including at least temperature range, vibration conditions and electromagnetic compatibility level. The parameter weight matrix is ​​generated by the hierarchical gated recurrent unit (HGRU), and its calculation formula is as follows: ； The semicolon ";" indicates a vector concatenation operation, which concatenates scene feature vectors. and environmental constraint vector The concatenated input vectors form a higher-dimensional joint input; the hierarchical gated recurrent unit (HGRU) performs temporal modeling and feature extraction on the concatenated input vectors to capture the coupling relationship between the application scenario and environmental conditions, and finally outputs a set of unnormalized weight representations. Softmax is used to transform an arbitrary real vector into a probability distribution, normalize the weights of the HGRU output, and make the sum of all parameter weights equal to 1. The final parameter weight matrix W corresponds to the importance of a technical parameter in each dimension, which includes at least temperature resistance, package compatibility, and electrical performance.

6. The component replacement selection method according to claim 5, characterized in that, The rule engine is a formal rule validation layer based on Datalog, used to define electronic constraints and industry standards. The rule execution logic is as follows: a candidate solution must pass all applicable rule validations; otherwise, the adaptive weight model will be triggered to redistribute weights or the candidate solution will be removed.

7. The component replacement selection method according to claim 1, characterized in that, In step S4, the supply chain collaborative optimization of the multi-objective optimization model includes the following steps: Step S4.1: Supply Chain Data Integration: Connect to the supplier database via RESTful API to obtain data including at least real-time inventory status, delivery date forecasts, and price volatility coefficients. Then, normalize and score this data using the following formula: ； ； Step S4.2: Construct a multi-objective optimization function to maximize the weighted sum of technology matching degree and supply chain stability. The objective function is: ； in, For technical compatibility, Rating the supply chain This is a scene adjustment coefficient, with a value of 0.7 in industrial scenarios and 0.5 in consumer electronics scenarios. Step S4.3: Apply preset constraints; the constraints are: technology matching degree. ≥0.8, delivery time score >0.6, inventory score >0.4; Step S4.4: Real-time filtering and sorting; the Pareto front screening algorithm is used to extract the non-dominated solution set, and the non-dominated solution with the highest technology matching degree is recommended first according to the output strategy, and the solution with the best supply chain stability is selected second.

8. A component substitution selection system for supply chain collaboration, characterized in that, The component substitution selection method using supply chain collaboration as described in any one of claims 1 to 7 includes: A dynamic knowledge graph management module is used to construct and dynamically update an electronic component knowledge graph based on multi-source heterogeneous data, and to obtain a candidate replacement set; the multi-source heterogeneous data includes at least component parameter data, supply chain operation data, and design environment data; this module includes: A multimodal data extraction unit is used to jointly parse unstructured data and extract key parameters; the unstructured data includes at least PDF datasheets, circuit schematics, and BOM tables; the extracted key parameters include at least structured parameters and unstructured relationships; the structured parameters include at least electrical characteristics and physical characteristics; the unstructured relationships are implicit relationships mined through graph attention networks; The incremental update and event response unit is used to respond to component state change events, incrementally update the state of the corresponding node and / or the association between nodes in the dynamic knowledge graph, and resolve parameter conflicts based on the confidence assessment of Bayesian inference. An adaptive similarity calculation engine is used to calculate the matching degree between each component and technical requirement in the candidate alternative set based on the target application scenario using a meta-learning-driven adaptive weight model, and to verify the effectiveness of the candidate solutions through an integrated rule engine. The supply chain collaborative optimization module is used to integrate real-time supply chain data and perform joint optimization scoring on verified candidate solutions through a multi-objective optimization model. The scoring combines technical matching degree and supply chain stability under preset constraints. The recommendation and feedback closed-loop module is used to output the optimized and ranked alternative selection recommendation results, and to iteratively update the dynamic knowledge graph and the adaptive weight model based on user feedback data.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor runs the program, it implements the component replacement selection method for supply chain collaboration as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions or computer programs, which, when processed and executed by a processor, implement the component substitution selection method for supply chain collaboration as described in any one of claims 1 to 7.

Images